setuptools_scm_git_archive seems suddenly broken

Replace broken link to c't article by permalink

.git_archival.txt

@@ -0,0 +1 @@
+ref-names: $Format:%D$

.gitattributes

@@ -5,4 +5,6 @@
 # (binary is a macro for -text -diff)
 *.jar	binary
 *.pdf	binary
-*.PDF	binary
+*.PDF	binary
+
+.git_archival.txt  export-subst

Dockerfile

@@ -23,12 +23,8 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 
 RUN apt-get install -y --no-install-recommends \
   unpaper \
-  libopenjpeg-dev \
-  ghostscript \
   qpdf \
-  poppler-utils
-
-RUN apt-get install -y --no-install-recommends \
+  poppler-utils \
   tesseract-ocr \
   tesseract-ocr-deu tesseract-ocr-spa tesseract-ocr-eng tesseract-ocr-fra
 
@@ -37,6 +33,14 @@ RUN apt-get install -qy --no-install-recommends \
   libpython3-dev \
   gcc
 
+# Install Ghostscript from Debian sid to work around JPEG 2000 issue in
+# Debian stretch libgs9 or gs 9.16~dfsg-2.1
+
+COPY ./share/etc-apt-sources.list /etc/apt/sources.list
+
+RUN apt-get update && apt-get install -y ghostscript/sid
+
+
 # Enforce UTF-8
 # Borrowed from https://index.docker.io/u/crosbymichael/python/ 
 RUN dpkg-reconfigure locales && \

README.rst

@@ -116,8 +116,8 @@ Install or upgrade the required Homebrew packages, if any are missing::
    brew install qpdf
    brew install ghostscript
    brew install python3
-   brew install libxml2
-   brew install leptonica
+   brew install libxml2 libffi leptonica
+   brew install unpaper    # optional
    brew install tesseract
    
 Update the homebrew pip and install Pillow::
@@ -252,11 +252,11 @@ In case you detect an issue, please:
 Press & Media
 -------------
 
--  `c't 1-2014, page 59 <http://www.heise.de/ct/inhalt/2014/1/58/>`__:
+-  `c't 1-2014, page 59 <http://heise.de/-2279695>`__:
    Detailed presentation of OCRmyPDF v1.0 in the leading German IT
    magazine c't
 -  `heise Open Source, 09/2014: Texterkennung mit
-   OCRmyPDF <http://www.heise.de/-2356670>`__
+   OCRmyPDF <http://heise.de/-2356670>`__
 
 Disclaimer
 ----------

RELEASE_NOTES.rst

@@ -6,6 +6,48 @@ Please always read this file before installing the package
 Download software here: https://github.com/jbarlow83/OCRmyPDF/tags
 
 
+v4.0.6:
+=======
+
+-  Update install instructions
+-  Provide a sRGB profile instead of using Ghostscript's
+
+
+v4.0.5:
+=======
+
+Fixes
+-----
+
+-  Remove some verbose debug messages from v4.0.4
+-  Fixed temporary that wasn't being deleted
+-  DPI is now calculated correctly for cropped images, along with other image transformations
+-  Inline images are now checked during DPI calculation instead of rejecting the image
+
+v4.0.4:
+=======
+
+Released with verbose debug message turned on. Do not use. Skip to v4.0.5.
+
+
+v4.0.3:
+=======
+
+New features
+------------
+
+-  Page orientations detected are now reported in a summary comment
+
+
+Fixes
+-----
+
+-  Show stack trace if unexpect errors occur
+-  Treat "too few characters" error message from Tesseract as a reason to skip that page rather than
+   abort the file
+-  Docker: fix blank JPEG2000 issue by insisting on Ghostscript versions that have this fixed
+
+
 v4.0.2:
 =======
 

ocrmypdf/ghostscript.py

@@ -5,6 +5,7 @@ from tempfile import NamedTemporaryFile
 from subprocess import Popen, PIPE, check_call
 from shutil import copy
 from . import get_program
+from .pdfa import SRGB_ICC_PROFILE
 
 
 def rasterize_pdf(input_file, output_file, xres, yres, raster_device, log,
@@ -52,7 +53,7 @@ def generate_pdfa(pdf_pages, output_file, threads=1):
             "-dJPEGQ=95",
             "-dPDFA=2",
             "-sPDFACompatibilityPolicy=2",
-            "-sOutputICCProfile=srgb.icc",
+            "-sOutputICCProfile=" + SRGB_ICC_PROFILE,
             "-sOutputFile=" + gs_pdf.name,
         ]
         args_gs.extend(pdf_pages)

ocrmypdf/main.py

@@ -378,7 +378,7 @@ def cleanup_working_files(*args):
 @transform(
     input=options.input_file,
     filter=formatter('(?i)\.pdf'),
-    output=work_folder + '{basename[0]}.repaired.pdf',
+    output=os.path.join(work_folder, '{basename[0]}.repaired.pdf'),
     extras=[_log, _pdfinfo, _pdfinfo_lock])
 def repair_pdf(
         input_file,
@@ -550,6 +550,12 @@ def orient_page(
         with open(output_file, 'wb') as out:
             writer.write(out)
 
+        with pdfinfo_lock:
+            pageno = int(os.path.basename(page_pdf)[0:6]) - 1
+            pageinfo = pdfinfo[pageno].copy()
+            pageinfo['rotated'] = orient_conf.angle
+            pdfinfo[pageno] = pageinfo
+
 
 @transform(
     input=orient_page,
@@ -1045,6 +1051,8 @@ def run_pipeline():
                 msg = "Error occurred while running this command:"
                 _log.error(msg + '\n' + exc_value)
                 return ExitCode.child_process_error
+            elif not options.verbose:
+                _log.error(e)
 
         return ExitCode.other_error
     except Exception as e:
@@ -1055,6 +1063,20 @@ def run_pipeline():
         _log.warning('Output file: The generated PDF/A file is INVALID')
         return ExitCode.invalid_output_pdfa
 
+    with _pdfinfo_lock:
+        _log.debug(_pdfinfo)
+        direction = {0: 'n', 90: 'e',
+                     180: 's', 270: 'w'}
+        orientations = []
+        for n, page in enumerate(_pdfinfo):
+            angle = _pdfinfo[n].get('rotated', 0)
+            if angle != 0:
+                orientations.append('{0}{1}'.format(
+                    n + 1,
+                    direction.get(angle, '')))
+        if orientations:
+            _log.info('Page orientations detected: ' + ' '.join(orientations))
+
     return ExitCode.ok
 
 

ocrmypdf/pageinfo.py

@@ -6,7 +6,9 @@ from decimal import Decimal, getcontext
 import re
 import sys
 import PyPDF2 as pypdf
+from collections import namedtuple
 
+matrix_mult = pypdf.pdf.utils.matrixMultiply
 
 FRIENDLY_COLORSPACE = {
     '/DeviceGray': 'gray',
@@ -19,7 +21,11 @@ FRIENDLY_COLORSPACE = {
     '/Indexed': 'index',
     '/Separation': 'sep',
     '/DeviceN': 'devn',
-    '/Pattern': '-'
+    '/Pattern': '-',
+    '/G': 'gray',  # Abbreviations permitted in inline images
+    '/RGB': 'rgb',
+    '/CMYK': 'cmyk',
+    '/I': 'index',
 }
 
 FRIENDLY_ENCODING = {
@@ -27,6 +33,8 @@ FRIENDLY_ENCODING = {
     '/DCTDecode': 'jpeg',
     '/JPXDecode': 'jpx',
     '/JBIG2Decode': 'jbig2',
+    '/CCF': 'ccitt',  # Abbreviations permitted in inline images
+    '/DCT': 'jpeg'
 }
 
 FRIENDLY_COMP = {
@@ -38,38 +46,159 @@ FRIENDLY_COMP = {
 }
 
 
-def _page_has_inline_images(page):
-    # PDF always uses \r\n for separator regardless of platform
-    # Really basic heuristic that might trigger the odd false positive
-    # This is only finds the first image and is not quite spec compliant
-    try:
-        contents = page.getContents()
-        data = contents.getData()
-    except AttributeError:
-        # If we can't access the contents or data (empty page?) then there
-        # are no inline images
-        return False
+def _matrix_from_shorthand(shorthand):
+    """Convert from PDF matrix shorthand to full matrix
 
-    begin_image, image_data, end_image = False, False, False
-    for data in re.split(b'\s+', data):
-        if data == b'BI':
-            begin_image = True
-        elif data == b'ID':
-            image_data = True
-        elif data == b'EI':
-            end_image = True
-        if all((begin_image, image_data, end_image)):
-            return True
-    return False
+    PDF 1.7 spec defines a shorthand for describing the entries of a matrix
+    since the last column is always (0, 0, 1).
+    """
+
+    a, b, c, d, e, f = map(float, shorthand)
+    return ((a, b, 0),
+            (c, d, 0),
+            (e, f, 1))
 
 
-def _find_page_images(page, pageinfo):
+def _shorthand_from_matrix(matrix):
+    """Convert from transformation matrix to PDF shorthand."""
+    a, b = matrix[0][0], matrix[0][1]
+    c, d = matrix[1][0], matrix[1][1]
+    e, f = matrix[2][0], matrix[2][1]
+    return tuple(map(float, (a, b, c, d, e, f)))
+
+
+def euclidean_distance(rowvec1, rowvec2):
+    return ((rowvec1[0] - rowvec2[0]) ** 2
+        + (rowvec1[1] - rowvec2[1]) ** 2) ** 0.5
+
+
+ContentsInfo = namedtuple('ContentsInfo',
+    ['raster_settings', 'inline_images'])
+
+def _interpret_contents(contentstream):
+    operations = contentstream.operations
+    stack = []
+    ctm = _matrix_from_shorthand((1, 0, 0, 1, 0, 0))
+    image_raster_settings = []
+    inline_images = []
+
+    for op in operations:
+        operands, command = op
+        if command == b'q':
+            stack.append(ctm)
+        elif command == b'Q':
+            ctm = stack.pop()
+        elif command == b'cm':
+            ctm = matrix_mult(
+                ctm, _matrix_from_shorthand(operands))
+        elif command == b'Do':
+            image_name = operands[0]
+            image_raster_settings.append(
+                (image_name, _shorthand_from_matrix(ctm)))
+        elif command == b'INLINE IMAGE':
+            settings = operands['settings']
+            inline_images.append(
+                (settings, _shorthand_from_matrix(ctm)))
+
+    return ContentsInfo(
+        raster_settings=image_raster_settings,
+        inline_images=inline_images)
+
+
+def _get_dpi(ctm_shorthand, image_size):
+    """Given the transformation matrix and image size, find the image DPI.
+
+    PDFs do not include image resolution information within image data.
+    Instead, the PDF page content stream describes the location where the
+    image will be rasterized, and the effective resolution is the ratio of the
+    pixel size to raster target size.
+
+    Normally a scanned PDF has the paper size set appropriately but this is
+    not guaranteed. The most common case is a cropped image will change the
+    page size (/CropBox) without altering the page content stream. That means
+    it is not sufficient to assume that the image fills the page, even though
+    that is the most common case.
+
+    This code solves the general case where the image may be scaled (always),
+    cropped, translated (often), and rotated in place (occasionally) to an
+    arbitrary angle (rare). It will work as long as the image is a
+    parallelogram from the perspective of a rectilinear coordinate system.
+    It does not work for arbitrarily quadrilaterals that might be produced
+    by shearing, but by that point DPI becomes a linear gradient rather than
+    constant over the image.
+
+    The transformation matrix describes the coordinate system at the time of
+    rendering. We transform the image corner locations into the coordinate
+    system and measure the width and height within the system, expressed in
+    PDF units. From there we can compare to the actual image dimensions.
+
+    pdfimages -list does calculate the DPI in some way that is not completely
+    naive, but it does not the DPI of rotated images right, so cannot be
+    used anymore to validate this. Photoshop works, or using Acrobat to
+    rotate the image back to normal.
+
+    It does not matter if the image is partially cropped, or even out of the
+    /MediaBox.
+
+    """
+    matrix = _matrix_from_shorthand(ctm_shorthand)
+
+    # Corners of the image in untransformed unit space; last
+    # column is a dummy to assist matrix math
+    corners = [[0, 0, 1],
+               [1, 0, 1],
+               [0, 1, 1],
+               [1, 1, 1]]
+
+    # Rotate/translate/scale the corners into PDF coords (1/72")
+    # ordering of points may change, e.g. if rotation is 180 then
+    # the point (0, 0) may become the top right
+    # The row vectors can all be transformed together here by building
+    # a matrix of them
+    page_unit_corners = matrix_mult(corners, matrix)
+
+    # Calculate the width and height of the rotated image
+    # the transformation matrix so the corner that was originally
+    # (1, 1) can be ignored
+    image_drawn_width = euclidean_distance(
+        page_unit_corners[0], page_unit_corners[1])
+    image_drawn_height = euclidean_distance(
+        page_unit_corners[0], page_unit_corners[2])
+
+    # print((image_drawn_width, image_drawn_height))
+
+    # The scale of the image is pixels per PDF unit (1/72")
+    scale_w = image_size[0] / image_drawn_width
+    scale_h = image_size[1] / image_drawn_height
+
+    # DPI = scale * 72
+    dpi_w = scale_w * 72.0
+    dpi_h = scale_h * 72.0
+
+    return (dpi_w, dpi_h)
+
+
+def _find_page_images(page, pageinfo, contentsinfo):
+
+    for n, im in enumerate(contentsinfo.inline_images):
+        settings, shorthand = im
+        image = {}
+        image['name'] = str('inline-%02d' % n)
+        image['width'] = settings['/W']
+        image['height'] = settings['/H']
+        image['bpc'] = settings['/BPC']
+        image['color'] = FRIENDLY_COLORSPACE.get(settings['/CS'], '-')
+        image['comp'] = FRIENDLY_COMP.get(image['color'], '?')
+
+        dpi_w, dpi_h = _get_dpi(shorthand, (image['width'], image['height']))
+        image['dpi_w'], image['dpi_h'] = Decimal(dpi_w), Decimal(dpi_h)
+        yield image
+
+    # Look for XObject (out of line images)
     try:
         page['/Resources']['/XObject']
     except KeyError:
         return
-
-    # Look for XObject (out of line images)
     for xobj in page['/Resources']['/XObject']:
         # PyPDF2 returns the keys as an iterator
         pdfimage = page['/Resources']['/XObject'][xobj]
@@ -79,6 +208,7 @@ def _find_page_images(page, pageinfo):
             if pdfimage['/ImageMask']:
                 continue
         image = {}
+        image['name'] = str(xobj)
         image['width'] = pdfimage['/Width']
         image['height'] = pdfimage['/Height']
         image['bpc'] = pdfimage['/BitsPerComponent']
@@ -98,8 +228,24 @@ def _find_page_images(page, pageinfo):
             image['color'] = 'jpx' if image['enc'] == 'jpx' else '?'
 
         image['comp'] = FRIENDLY_COMP.get(image['color'], '?')
-        image['dpi_w'] = image['width'] / pageinfo['width_inches']
-        image['dpi_h'] = image['height'] / pageinfo['height_inches']
+        image['dpi_w'] = image['dpi_h'] = 0
+
+        for raster in contentsinfo.raster_settings:
+            # Loop in case the same image is display multiple times on a page
+            if raster[0] != image['name']:
+                continue
+            shorthand = raster[1]
+
+            dpi_w, dpi_h = _get_dpi(
+                shorthand, (image['width'], image['height']))
+
+            # When image is used multiple times take the highest DPI it is
+            # rendered at
+            image['dpi_w'] = max(dpi_w, image.get('dpi_w', 0))
+            image['dpi_h'] = max(dpi_h, image.get('dpi_h', 0))
+
+        image['dpi_w'] = Decimal(image['dpi_w'])
+        image['dpi_h'] = Decimal(image['dpi_h'])
         image['dpi'] = (image['dpi_w'] * image['dpi_h']) ** Decimal(0.5)
         yield image
 
@@ -141,12 +287,14 @@ def _pdf_get_pageinfo(infile, pageno: int):
     pageinfo['width_inches'] = width_pt / Decimal(72.0)
     pageinfo['height_inches'] = height_pt / Decimal(72.0)
 
-    pageinfo['images'] = [im for im in _find_page_images(page, pageinfo)]
+    try:
+        contentstream = pypdf.pdf.ContentStream(page.getContents(), pdf)
+    except AttributeError as e:
+        return pageinfo
 
-    # Look for inline images
-    if _page_has_inline_images(page):
-        raise NotImplementedError(
-            "Warning: input PDF contains inline images - not supported")
+    contentsinfo = _interpret_contents(contentstream)
+    pageinfo['images'] = [im for im in _find_page_images(
+                                page, pageinfo, contentsinfo)]
 
     if pageinfo['images']:
         xres = max(image['dpi_w'] for image in pageinfo['images'])
@@ -164,3 +312,17 @@ def pdf_get_all_pageinfo(infile):
     pdf = pypdf.PdfFileReader(infile)
     getcontext().prec = 6
     return [_pdf_get_pageinfo(infile, n) for n in range(pdf.numPages)]
+
+
+def main():
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument('infile')
+    args = parser.parse_args()
+    info = pdf_get_all_pageinfo(args.infile)
+    from pprint import pprint
+    pprint(info)
+
+
+if __name__ == '__main__':
+    main()

ocrmypdf/pdfa.py

@@ -5,10 +5,13 @@
 
 from __future__ import print_function, absolute_import, division
 from string import Template
-from subprocess import Popen, PIPE
-import os
 import codecs
-from . import get_program
+import pkg_resources
+
+ICC_PROFILE_RELPATH = 'data/sRGB_IEC61966-2-1_black_scaled.icc'
+
+SRGB_ICC_PROFILE = pkg_resources.resource_filename(
+    'ocrmypdf', ICC_PROFILE_RELPATH)
 
 
 # This is a template written in PostScript which is needed to create PDF/A
@@ -93,38 +96,9 @@ def _get_pdfa_def(icc_profile, icc_identifier, pdfmark):
     return result
 
 
-def _get_postscript_icc_path():
-    "Parse Ghostscript's help message to find where iccprofiles are stored"
-
-    p_gs = Popen([get_program('gs'), '--help'], close_fds=True,
-                 universal_newlines=True,
-                 stdout=PIPE, stderr=PIPE)
-    out, _ = p_gs.communicate()
-    lines = out.splitlines()
-
-    def search_paths(lines):
-        seeking = True
-        for line in lines:
-            if seeking:
-                if line.startswith('Search path'):
-                    seeking = False
-                    continue
-            else:
-                if line.strip().startswith('/'):
-                    yield from (
-                        path.strip() for path in line.split(':')
-                        if path.strip() != '')
-    for root in search_paths(lines):
-        path = os.path.realpath(os.path.join(root, '../iccprofiles'))
-        if os.path.exists(path):
-            return path
-
-    raise FileNotFoundError("Could not find Ghostscript's iccprofiles")
-
-
 def generate_pdfa_def(target_filename, pdfmark, icc='sRGB'):
     if icc == 'sRGB':
-        icc_profile = os.path.join(_get_postscript_icc_path(), 'srgb.icc')
+        icc_profile = SRGB_ICC_PROFILE
     else:
         raise NotImplementedError("Only supporting sRGB")
 

ocrmypdf/tesseract.py

@@ -98,7 +98,8 @@ def get_orientation(input_file, language: list, timeout: float, log):
         return OrientationConfidence(angle=0, confidence=0.0)
     except CalledProcessError as e:
         tesseract_log_output(log, e.output, input_file)
-        if 'Image too large' in e.output:
+        if ('Too few characters. Skipping this page' in e.output or
+                'Image too large' in e.output):
             return OrientationConfidence(0, 0)
         raise e from e
     else:

setup.py

@@ -209,7 +209,6 @@ setup(
         ],
     setup_requires=[
         'setuptools_scm',
-        'setuptools_scm_git_archive',
         'cffi>=1.5.0',
         'pytest-runner'
     ],
@@ -231,5 +230,6 @@ setup(
             'ocrmypdf = ocrmypdf.main:run_pipeline'
         ],
     },
+    package_data={'ocrmypdf': ['data/sRGB_IEC61966-2-1_black_scaled.icc']},
     include_package_data=True,
     zip_safe=False)

share/etc-apt-sources.list

@@ -0,0 +1,4 @@
+deb http://httpredir.debian.org/debian stretch main
+deb http://httpredir.debian.org/debian stretch-updates main
+deb http://security.debian.org stretch/updates main
+deb http://ftp.de.debian.org/debian sid main contrib non-free

tests/test_pageinfo.py

@@ -103,8 +103,8 @@ def test_single_page_image():
     assert pdfimage['bpc'] == 8
 
     # DPI in a 1"x1" is the image width
-    assert pdfimage['dpi_w'] == 8
-    assert pdfimage['dpi_h'] == 8
+    assert abs(pdfimage['dpi_w'] - 8) < 1e-5
+    assert abs(pdfimage['dpi_h'] - 8) < 1e-5
 
 
 def test_single_page_inline_image():
@@ -120,8 +120,12 @@ def test_single_page_inline_image():
         pdf.showPage()
         pdf.save()
 
-    with pytest.raises(NotImplementedError):
-        pageinfo.pdf_get_all_pageinfo(filename)
+    pdfinfo = pageinfo.pdf_get_all_pageinfo(filename)
+    print(pdfinfo)
+    pdfimage = pdfinfo[0]['images'][0]
+    assert (pdfimage['dpi_w'] - 8) < 1e-5
+    assert pdfimage['color'] != '-'
+    assert pdfimage['width'] == 8
 
 
 def test_jpeg():
@@ -131,4 +135,5 @@ def test_jpeg():
 
     pdfimage = pdfinfo[0]['images'][0]
     assert pdfimage['enc'] == 'jpeg'
+    assert (pdfimage['dpi_w'] - 150) < 1e-5